Abstract and Introduction
Pharmacoeconomic guidelines have focused on improving the measurement of effectiveness, but have largely ignored the assessment of cost inputs. We explored differences in adverse event cost estimates derived from three different sources in a case study of selective COX2 NSAIDs. We found substantial differences in costs of adverse events across data sources. We also determined that treatment costs associated with adverse events differed substantially based upon inclusion/exclusion criteria. Health services researchers should be deliberate in the choice of adverse event treatment costs that are used in decision analytic models. Future guidelines should seek to delineate best practices for cost calculations.
One of the first major efforts towards the standardization of pharmacoeconomics methodology in the USA came in 1996, with the publication of a detailed guidance from the Panel on Cost–Effectiveness in Health and Medicine. The Panel provided the first set of specific methodological recommendations for the development of cost–effectiveness analyses, and the guidance is still considered to be one of the gold standards in the field today. Since the publication of the Panel's report, several other organizations have developed additional recommendations for economic evaluation. One notable example is the Format for Formulary Submissions, issued by the Academy of Managed Care Pharmacy (AMCP). The AMCP Format provides guidance on how to standardize pharmacoeconomic modeling and reporting of information for submission to formulary decision-makers.
In the last 5 years, several other reports have been issued by the International Society for Pharmacoeconomic and Outcomes Research (ISPOR) that have focused on improving specific aspects of methods used in pharmacoeconomic analyses and real-world studies. For example, ISPOR has issued best-practice recommendations on the development of economic analyses based on clinical trial data as well as analyses based on retrospective databases. More recently, the organization issued a guidance on the use of real-world (i.e., observational) data.
Taken as a whole, efforts such as these are designed to standardize the techniques used to estimate treatment effectiveness, and to encourage health services researchers to calculate effectiveness measures in a way that is both accurate and relevant to the decision-maker. As pharmacoeconomic methods have become increasingly transparent and standardized, decision-makers have become more comfortable reviewing cost–effectiveness analyses. As a result, such analyses are more frequently incorporated into health technology assessments.
To date, the primary focus of the guidelines related to pharmacoeconomic analyses has been on three areas: methods for obtaining unbiased estimates of treatment effectiveness, best practices for calculating incremental cost–effectiveness ratios (ICERs), and the appropriate use of sensitivity analysis. There has been less emphasis placed on the methods for deriving cost inputs as components of pharmacoeconomic studies. Although not explicitly stated, pharmacoeconomic guidelines have long made the assumption that costs for primary end points and adverse events could be obtained from payers. However, it is often the case that these cost estimates are not readily available, or that if available, these values are proprietary.
Input costs are a critical component of pharmacoeconomic models, yet there is no standard methodology for estimating such costs, and pharmacoeconomic studies of similar treatments often use costs from a variety of divergent sources, making comparison of results difficult. Furthermore, the question of which populations should be examined to ascertain costs has not been adequately addressed in the literature.
To better understand the importance of accurate cost inputs in the context of pharmacoeconomic analyses, consider the following adaptation of the ICER formula:
and ΔCost represents the true difference in costs between the two treatments under consideration; ΔEffectiveness represents the true difference in effectiveness between the treatments; λ represents the factor of error surrounding the observation of costs; and Ψ represents the factor of error surrounding the observation of effectiveness.
Let us assume that the estimation of cost and effectiveness is not perfect; the values of λ and Ψ will not equal 1, and any deviation away from 1 represents bias in the numerator and denominator of the ICER for λ and Ψ, respectively. Thus, λ and Ψ indicate how far the estimates are from the ‘true' cost–effectiveness ratio. Assuming that the measure of effectiveness has been correctly calculated, the value of Ψ will be close to 1. If the methods used to derive costs provide an estimate of ΔCost that is significantly higher than the true value, then λ will be significantly greater than 1, and the ratio will be much greater than the true value. Conversely, if the costing methods provide an underestimate of the true difference in costs, then the estimated ICER will be smaller than the true value.
Sensitivity analysis enables the testing of assumptions, but should not be used to address known errors in estimation; if there are known errors in estimation, a correction to address these known errors should be performed when calculating the baseline estimates rather than in sensitivity analysis. If the baseline ICER is biased, any sensitivity analysis also will be biased. If λ were known to be different from 1 and Ψ were assumed to equal to 1, then the best unbiased estimate of the ICER would be ICEREstimated /λ. Continuing to assume that Ψ is 1, if the variance of the effectiveness estimate is large (i.e., the standard error surrounding Ψ were large), then as sensitivity analyses of the effectiveness parameter are performed, the bias of λ would be exacerbated.
Accuracy of estimation is critical in order for ICERs to be useful for decision-makers. The appropriate estimation of effectiveness has long been a concern in cost–effectiveness analyses, but the appropriate estimation of cost is equally important. Cost estimates based upon general population parameters, even considering sensitivity analyses, probably do not reflect the true costs for more specific populations; namely, the populations likely to be prescribed the drug of interest. As such, a cost–effectiveness ratio derived from using general population input costs could be markedly different from the cost–effectiveness ratio derived from using the target population for ascertainment of input costs.
In order to explore the potential impact that different costing approaches might have on the results of a pharmacoeconomic analyses, we analyzed cost components that would be used in an economic assessment that compared a selective COX2 inhibiting NSAID against a nonselective NSAID. Following their introduction in the late 1990s, the selective COX2s were a popular choice for the treatment of chronic pain. This new class of pain medications offered efficacy similar to that of nonselective NSAIDs, with the added benefit of fewer gastrointestinal (GI) side effects. Utilization of the three marketed COX2 inhibitors (celecoxib, rofecoxib and valdecoxib) grew steadily until 2004, when reports suggested that the use of certain selective COX2 agents might be associated with an increased risk of myocardial infarction and stroke. The evolving evidence base for cardiovascular risk and other safety concerns eventually led to the voluntary withdrawals of rofecoxib (Vioxx®) and valdecoxib (Bextra®), and the addition of a black-box warning for celecoxib (Celebrex®).
Since the reports on the potential cardiovascular safety concerns related to the use of selective COX2 inhibitors first appeared, there has been substantial debate in the medical community regarding the nature of increased risk associated with the selective COX2 NSAIDs.[5–11] Some have argued that the selective COX2s represent an effective treatment alternative to the nonselective NSAIDs for patients experiencing severe chronic pain, assuming that the potential risks associated with treatment are well understood by both the patient and the clinician. Proponents of selective NSAIDs also point out that the use of nonselective NSAIDs is associated with an increased risk of serious GI complications. Others suggest that the risk/benefit profile of selective COX2s would not outweigh the risk/benefit profile of the nonselective NSAIDs, and that the traditional treatment alternatives must be considered.
For the purposes of this analysis, the inpatient cost to treat a series of adverse events that would be included in a pharmacoeconomic model of a COX2 NSAID were considered. Specifically, the cost to treat complicated GI, hepatic, renal failure, symptomatic ulcer and cardiovascular (Antiplatelet Trialists' Collaboration [APTC] end point) events among patients receiving COX2 NSAIDS, traditional NSAIDS and no NSAIDs were assessed using data from three different sources that are commonly used in pharmacoeconomic analyses:
Insurance claims data
Medicare reimbursement data
Our goal was to compare and contrast the cost estimates obtained from each of the three sources, and to gauge the impact the use of data from different cost sources may have on the results of a cost–effectiveness analysis.
Expert Rev Pharmacoeconomics Outcomes Res. 2008;8(5):463-469. © 2008 Expert Reviews Ltd.
No writing assistance was utilized in the production of this review manuscript.
Cite this: Impact of Patient Sample on Costs of Events in Pharmacoeconomic Models - Medscape - Oct 01, 2008.